Jaepil Ko

A Simple Illumination Normalization Algorithm for Face Recognition

Most of the FR (face recognition) systems suffer from sensitivity to variations in illumination. For better performance the FR system needs more training samples acquired under variable lightings but it is not practical in real world. We introduce a novel pre-processing method, which makes illumination-normalized face image for face recognition. The proposed method, ICR (Illumination Compensation based on Multiple Regression Model), is to find the plane that best fits the intensity distribution of the face image using the multiple regression model, then use this plane to normalize the face image. The advantages of our method are simple and practical. The planar approximation of a face image is mathematically defined by the simple linear model. We provide experimental results to demonstrate the performance of the proposed ICR method on public face databases and our database. The experiments show a significant improvement of the recognition rate.

Binary classifier fusion based on the basic decomposition methods

For a complex multiclass problem, it is common to construct the multiclass classifier by combining the outputs of several binary ones. The two basic methods for this purpose are known as one-per-class (OPC) and pairwise coupling (PWC) and their general form is error correcting output code (ECOC). In this paper, we review basic decomposition methods and introduce a new sequential fusion method based on OPC and PWC according to their properties. In the experiments, we compare our proposed method with each basic method and ECOC method. The experimental results show that our proposed method can improve significantly the classification accuracy on the real dataset.

Improved N-division output coding for multiclass learning problems

The output coding for multiclass learning problems is a generalization of one-per-class, all-pairs, and error correcting output codes. Although, the prevailing concepts of output coding have been error correcting properties, the one-per-class and all-pairs are still considered to be one of the state-of-art methods. However, these two methods are contrary to each other in the aspect of producing complex dichotomies and the problem of nonsense outputs. In additions, they all perform a prior decomposition without regards to the properties of a given training data set. In this paper, we propose a new data-driven output coding method that is the generalized form of one-per-class and all-pairs. We present the properties of the proposed method. From experimental results on both a toy problem and real benchmark datasets, we present that our proposed method achieves a comparable performance with good properties.

N -division output coding method applied to face recognition

Most research on face recognition has focused on representation of face appearances rather than the classifiers. For robust classification performance, we need to adopt elaborate classifiers. Output coding is suitable for this purpose because it can allow online learning. In this paper, we propose an N-division output coding method. In the experiments we demonstrate such properties as problem complexity, margin of separation, machine relevance and the recognition performance among different output coding methods.

Combining SVM classifiers for multiclass problem: its application to face recognition

In face recognition, a simple classifier such as k-NN is frequently used. For a robust system, it is common to construct the multi-class classifier by combining the outputs of several binary ones. The two basic schemes for this purpose are known as one-per-class (OPC) and pairwise coupling (PWC). The performance of decomposition methods depends on accuracy of base dichotomizers. Support vector machine is suitable for this purpose. In this paper, we give the strength and weakness of two representative decomposition methods, OPC and PWC. We also introduce a new method combining OPC and PWC with rejection based on the analysis of OPC and PWC using SVM as base classifiers. The experimental results on the ORL face database show that our proposed method can reduce the error rate on the real dataset.

Various decomposition methods applied to face recognition

Face recognition has mainly focused on face representation, so a simple classifier is frequently used. For a robust system, it is common to construct a multiclass classifier by combining outputs of several binary ones. In this paper, we overviews basic decomposition and decoding schemes and propose new methods then give empirical results of recognition performance on the ORL face dataset.

Multi-class Support Vector Machines with Case-Based Combination for Face Recognition

The support vector machine is basically to deal with a two-class classification problem. To get M-class classifiers for face recognition, it is common to construct a set of binary classifiers f 1...f m , each trained to separate one class from the rest. The multi-class classification method has a main shortcoming that the binary classifiers used are obtained by training on different binary classification problems, and thus it is unclear whether their real-valued outputs are on comparable scales. In this paper, we try to use additional information, relative outputs of the machines, for final decision. We propose case-based combination with reject option to use the information. The experiments on the ORL face database shows that the proposed method achieves a slight better performance than the previous multi-class support vector machines.

Illumination Normalized Face Image for Face Recognition

A small change in illumination produces large changes in appearance of face even when viewed in fixed pose. It makes face recognition more difficult to handle. To deal with this problem, we introduce a simple and practical method based on the multiple regression model, we call it ICR (Illumination Compensation based on the Multiple Regression Model). We can get the illumination-normalized image of an input image by ICR. To show the improvement of recognition performance with ICR, we applied ICR as a preprocessing step. We achieved better result with the method in preprocessing point of view when we used a popular technique, PCA, on a public database and our database.

Empirical remarks on output coding methods for face recognition

Since facial images are affected by various factors, the representation capacity for face database is limited by the prototypes collected for training. Therefore, to extend the capacity covering variations of facial images, we should adopt a complex classifier. It is desirable to use output coding method by considering the number of classes changes. We propose new output coding methods and then compare them with representative conventional output coding methods to investigate the properties of decomposition schemes through the experiment on the ORL face dataset. Finally, we give discussions on some factors that should be considered in the designing of decomposition scheme, to provide some foundation for designing new output coding methods in face recognition.